Due to software and hardware limitations, there are a couple of
limitations the profiling and analysis performed by the Visual
Profiler.
- The Multiprocessor, Kernel
Memory, and Kernel
Instruction analysis stages require metrics that
are only available on devices with compute capacity 2.0 or
higher. When these analyses are attempted on a device with a
compute capacity of 1.x, the analysis results will show that
the required data is "not available".
- The Kernel Instruction analysis stage
that determines branch divergence requires a metric that in not
available on 3.0 devices. When this analysis is attempted on a
device with a compute capacity of 3.0 the analysis results
will show that the required data is "not available".
- For a device, only a single context can be profiled during
the execution of the application. When metrics or events are
collected for an application that creates multiple contexts on
a device, those metrics and events will only be collected for
one of the contexts. In the analysis results, the
"insufficient data" message will be displayed to indicate that
there are one or more contexts for which analysis results
could not be collected due to this limitation.
- Some metric values are calculated assuming a kernel is large
enough to occupy all device multiprocessors with approximately
the same amount of work. If a kernel launch does not have this
characteristic, then those metric values may not be
accurate.
- Concurrent kernel execution is serialized when an
application is profiled. Thus the timeline will not display
concurrent kernel execution even if it occurs in the
application when running normally.